On 2/7/07, James Richard Tyrer <[EMAIL PROTECTED]> wrote:
No actually I meant swap, but what I said isn't very clear. I didn't mean adding _more_ swap will improve performance but rather that even with a lot of main memory that having swap (as contrasted with *no* swap) will improve performance. The reason is that stuff in application memory currently not being used will be swapped out making more room for disk cache and buffers. OTHO, the question of just how much swap is desirable is another question.
Of course, it depends on workload, and amount of available RAM as well... If my system constantly touches most of what it has in RAM, and happens to have several GB of it, swapping out a few tens of megabytes to make room for more cached files may get me an extra .5% performance, which may be negated the first time the system realizes it need to access something in the 40Mb of stuff in swap. Bottom line, it depends on your workload, your hardware, and the kernel's VM algorithms, but swap generally can increase performance. The performance increase allowed for by swap (and thus it's usefulness - fewer and fewer folks actually _need_ more ram than they have in order to avoid out-of-memory situations), is tending to decrease as RAM sizes increase. For my servers (with 3+ Gb or RAM), and my workload, the difference in performance afforded by swap is well within the margins of statistical error. I suspect that trend will continue, with swap performance increases tending toward 0 for an ever increasing segment of the computing population. Moore's law and all. --tim _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
